How to Check if a Source Really Exists (DOI, Crossref, Catalogues)

To check whether a source actually exists, you need one thing: a persistent identifier that resolves in a database nobody involved in your paper controls. In practice that means three moves — paste the DOI into doi.org, search the exact title in Crossref, and if neither returns anything, check a library catalogue or the publisher's own site. If none of the three confirms the item, you are not looking at an obscure source. You are looking at one that probably does not exist.

This is a step-by-step procedure. It also covers what a missing DOI does not prove, because that is the most common mistake people make when auditing a reference list: assuming anything without a DOI is fabricated.

Step 1: resolve the DOI

A DOI (digital object identifier) is not a link. It is an identifier defined by the ISO 26324:2012 standard, with a fixed and recognisable shape: a prefix that begins with the numeral 10., assigned by the International DOI Foundation or by a registration agency — Crossref, in the case of scholarly publishers — followed by a suffix the publisher assigns itself (Boudry and Chartron, 2017, p. 3). That gives you a free first test: a genuine DOI always starts with 10. and contains a slash. A string like DOI: 45.9921/xyz is invented before you check anything at all.

Real verification means pasting the identifier into https://doi.org/. The system works by having the publisher register the object's address, so incoming requests for a DOI are resolved to the appropriate URL (Boudry and Chartron, 2017, p. 3). Three things can happen:

• You land on the article's page at the publisher. The source exists. Now confirm that the title, authors and year match what sits in your reference list.
• You land on a real article, but a different one. The DOI is genuine and has been attached to somebody else's metadata. This is the signature failure mode of citations produced by a language model: a valid identifier paired with an invented title.
• You get "DOI not found". The identifier was never registered.

The same identifier can also return machine-readable metadata rather than a web page: an application can resolve a DOI while specifying acceptable formats in the Accept header of the HTTP request, and the resolver will return the matching resource instead of the usual landing page (Harvey et al., 2015, p. 3). That mechanism is what every automated bibliography checker is built on, including ours.

Step 2: what to do when there is no DOI

A missing DOI is not evidence of fabrication. Boudry and Chartron examined articles indexed in PubMed and found that 86.42% of the 2015 cohort carried a DOI (2017, p. 2) — meaning roughly one in seven real, indexed biomedical papers did not. They also found that adoption varies sharply by country: some gradually implemented DOIs across the period studied, while others were still assigning very few as recently as 2015 (2017, p. 2). On top of that, whole categories of legitimate scholarship routinely lack DOIs: older monographs, edited-volume chapters, government reports, dissertations, and many regional journals.

So for items without a DOI, drop down a level:

1. Crossref. Search the exact title in quotation marks. Crossref is the registration agency for scholarly publishers, so it covers a very large share of the journal literature.
2. A library catalogue — WorldCat, your university's own catalogue, or a national union catalogue. For books, the ISBN plays the role the DOI plays for articles: it is either in the catalogue or it is not.
3. The publisher's or the institution's repository, for reports, working papers and conference proceedings.

DOIs are also not the only persistent identifiers in circulation. URN and PURL sit alongside them, differ in how they resolve, and the choice between schemes depends on the publication type and the publisher's infrastructure (Paliulis and Dagienė, 2009, p. 159). If a reference carries an identifier that is not in 10.xxxx/... form, don't discard it — work out which system it belongs to first.

Step 3: learn the shape of a fabricated reference

Invented citations repeat a small set of patterns. Knowing them tells you which entries to check first:

• A plausible title that does not exist. It reads like a summary of your own argument, because it was assembled from the keywords in your paragraph.
• A real author in the wrong journal. A name that is genuinely prominent in the field, attached to a periodical they have never published in.
• Suspiciously tidy details. Volume 1, issue 1, pages 1–10, a very recent year.
• A DOI that resolves to something else, as described above.
• No trace anywhere but your own bibliography. A genuine paper that appears exactly once, in your reference list, and nowhere else online.

No single signal settles it. Two or three together effectively do.

Checking a whole reference list at once

By hand, a single entry takes two to four minutes. On a 40-item reference list for a dissertation, that is roughly ninety minutes of clicking, usually the night before submission. That is why we built the bibliography checker: paste in the whole list and it queries identifier databases, then flags which entries are confirmed, which have mismatched metadata, and which cannot be found anywhere.

When an entry really does turn out not to exist, you need a replacement — a genuine publication that supports the same sentence. That is what the source finder is for: paste the paragraph, get back real publications with page numbers you can cite.

Worth noting: every scholarly work cited above was retrieved by cytado itself, from its own source corpus, and the page numbers come from the pagination of the original files. It would be a strange thing to write about verifying citations while inventing our own.

In short

Verification is a sequence, not a single click: resolve the DOI → search Crossref → check a library catalogue. Confirmation at any one level is enough. Failure at all three means the entry should not stay in your reference list — however neatly it happens to fit the argument.